Paste your MCP server's tools/list and check it for prompt injection, tool poisoning, data-exfiltration and hidden-unicode payloads — free, in your browser, before you ship it or connect an agent.
Paste the array, or the whole JSON-RPC tools/list response. It is sent once to a read-only analysis endpoint that never stores it and never calls your tools — so a private or localhost server stays private. Load a poisoned example →
Every advertised tool's name, description and JSON schema is statically analysed for concealed instructions (including contraction-phrased "don't tell the user" variants), data-exfiltration hints, dangerous capabilities, cross-tool shadowing (one server hijacking another's tools), and invisible-unicode payloads — the OWASP LLM01 (prompt injection) and LLM08 (excessive agency) classes. Read the tool-poisoning explainer for how the attack works, or browse independent scans of popular public servers.
This preview gives you the verdict, risk score and severity counts. The paid audit adds which tool is flagged, the exact offending text, the OWASP mapping and remediation — GET /scan/mcp?url=… or zero-egress POST /scan/mcp with your tools array ($0.05 USDC via x402, no account). Running it on every build? A $9.99/mo plan gives an API key with 100 scans/month to gate CI on the verdict — pay by card, no wallet.